DeepSeek warns of a 2–10x API price hike, ending its price war
The cheapest frontier API cools after 8T tokens/day overwhelmed it — plus a free Ling 3.0 Tiny on Vercel, Ai2's 2PB open catalog, and Copilot review dials.
Copy markdown
Even a 10x hike stays cheapest
DeepSeek warned of a “significant” API price increase and told developers to “plan your usage accordingly”; founder Jun Song says even a 2–10x jump would keep it under OpenAI and Anthropic. V4-Flash still runs about $0.14 per million input tokens and $0.28 output — with no new rate or date published yet, now’s the time to add fallback providers and stress-test budgets.
Why the cheap era cracked
The tipping point: V4-Flash processed 8 trillion tokens in a single day on Aug 1, outrunning DeepSeek’s compute. It’s also floating peak-hour surge pricing during weekday business hours (Beijing time) — a break from years of undercutting rivals on razor-thin margins.
A free, tiny fallback in one line
inclusionAI’s Ling 3.0 Tiny just landed on Vercel AI Gateway with BYOK and automatic failover across API formats — and it’s free through mid-August. If DeepSeek’s bill worries you, it’s a drop-in cheap route to test this week.
Ai2's open catalog gets a bigger pipe
Ai2 tripled its Hugging Face storage to nearly 2 PB with no download rate limits, opening full-throughput pulls of ~900 models and 1,200+ datasets — Olmo, Molmo, OlmoEarth, MolmoAct — plus LeRobot integration for robotics. The self-host escape hatch from rising API bills just got easier to use.
Update: Copilot review effort goes GA
GitHub Copilot’s code-review effort levels are now generally available: dial a PR review between lighter and deeper passes to trade thoroughness for speed and cost. Previously in preview, it’s now live across surfaces.